* [PATCH, RFC] aic7xxx: semaphore to completion conversion
@ 2006-01-31 17:20 Christoph Hellwig
2006-02-06 14:40 ` Christoph Hellwig
0 siblings, 1 reply; 3+ messages in thread
From: Christoph Hellwig @ 2006-01-31 17:20 UTC (permalink / raw)
To: jejb, hare; +Cc: linux-scsi
switch eh_sem to a completion. due to wait_for_completion_timeout this
also nicely simplifies the code. Unfortunately it's untested, so if
someone with the hardware could give it a try that would be nice. Once
it works the same thing can be applied to aic79xx.
Index: scsi-misc-2.6/drivers/scsi/aic7xxx/aic7xxx_osm.c
===================================================================
--- scsi-misc-2.6.orig/drivers/scsi/aic7xxx/aic7xxx_osm.c 2006-01-13 17:55:57.000000000 +0100
+++ scsi-misc-2.6/drivers/scsi/aic7xxx/aic7xxx_osm.c 2006-01-31 17:28:31.000000000 +0100
@@ -373,7 +373,6 @@
struct scb *);
static void ahc_linux_queue_cmd_complete(struct ahc_softc *ahc,
struct scsi_cmnd *cmd);
-static void ahc_linux_sem_timeout(u_long arg);
static void ahc_linux_freeze_simq(struct ahc_softc *ahc);
static void ahc_linux_release_simq(struct ahc_softc *ahc);
static int ahc_linux_queue_recovery_cmd(struct scsi_cmnd *cmd, scb_flag flag);
@@ -1193,7 +1192,7 @@
memset(ahc->platform_data, 0, sizeof(struct ahc_platform_data));
ahc->platform_data->irq = AHC_LINUX_NOIRQ;
ahc_lockinit(ahc);
- init_MUTEX_LOCKED(&ahc->platform_data->eh_sem);
+ init_completion(&ahc->platform_data->eh_done);
ahc->seltime = (aic7xxx_seltime & 0x3) << 4;
ahc->seltime_b = (aic7xxx_seltime & 0x3) << 4;
if (aic7xxx_pci_parity == 0)
@@ -1832,7 +1831,7 @@
ahc_set_transaction_status(scb, CAM_CMD_TIMEOUT);
if ((ahc->platform_data->flags & AHC_UP_EH_SEMAPHORE) != 0) {
ahc->platform_data->flags &= ~AHC_UP_EH_SEMAPHORE;
- up(&ahc->platform_data->eh_sem);
+ complete(&ahc->platform_data->eh_done);
}
}
@@ -2040,22 +2039,6 @@
}
static void
-ahc_linux_sem_timeout(u_long arg)
-{
- struct ahc_softc *ahc;
- u_long s;
-
- ahc = (struct ahc_softc *)arg;
-
- ahc_lock(ahc, &s);
- if ((ahc->platform_data->flags & AHC_UP_EH_SEMAPHORE) != 0) {
- ahc->platform_data->flags &= ~AHC_UP_EH_SEMAPHORE;
- up(&ahc->platform_data->eh_sem);
- }
- ahc_unlock(ahc, &s);
-}
-
-static void
ahc_linux_freeze_simq(struct ahc_softc *ahc)
{
unsigned long s;
@@ -2355,25 +2338,16 @@
if (paused)
ahc_unpause(ahc);
if (wait) {
- struct timer_list timer;
- int ret;
-
ahc->platform_data->flags |= AHC_UP_EH_SEMAPHORE;
ahc_unlock(ahc, &flags);
- init_timer(&timer);
- timer.data = (u_long)ahc;
- timer.expires = jiffies + (5 * HZ);
- timer.function = ahc_linux_sem_timeout;
- add_timer(&timer);
printf("Recovery code sleeping\n");
- down(&ahc->platform_data->eh_sem);
- printf("Recovery code awake\n");
- ret = del_timer_sync(&timer);
- if (ret == 0) {
+ if (!wait_for_completion_timeout(
+ &ahc->platform_data->eh_done, 5 * HZ)) {
printf("Timer Expired\n");
retval = FAILED;
}
+ printf("Recovery code awake\n");
} else
ahc_unlock(ahc, &flags);
return (retval);
Index: scsi-misc-2.6/drivers/scsi/aic7xxx/aic7xxx_osm.h
===================================================================
--- scsi-misc-2.6.orig/drivers/scsi/aic7xxx/aic7xxx_osm.h 2006-01-13 17:55:57.000000000 +0100
+++ scsi-misc-2.6/drivers/scsi/aic7xxx/aic7xxx_osm.h 2006-01-31 17:28:42.000000000 +0100
@@ -369,7 +369,7 @@
spinlock_t spin_lock;
u_int qfrozen;
- struct semaphore eh_sem;
+ struct completion eh_done;
struct Scsi_Host *host; /* pointer to scsi host */
#define AHC_LINUX_NOIRQ ((uint32_t)~0)
uint32_t irq; /* IRQ for this adapter */
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH, RFC] aic7xxx: semaphore to completion conversion
2006-01-31 17:20 [PATCH, RFC] aic7xxx: semaphore to completion conversion Christoph Hellwig
@ 2006-02-06 14:40 ` Christoph Hellwig
0 siblings, 0 replies; 3+ messages in thread
From: Christoph Hellwig @ 2006-02-06 14:40 UTC (permalink / raw)
To: jejb, hare, emmanuel.fuste; +Cc: linux-scsi
On Tue, Jan 31, 2006 at 06:20:18PM +0100, Christoph Hellwig wrote:
> switch eh_sem to a completion. due to wait_for_completion_timeout this
> also nicely simplifies the code. Unfortunately it's untested, so if
> someone with the hardware could give it a try that would be nice. Once
> it works the same thing can be applied to aic79xx.
New version that switches to the common onstack completion and just a
pointer in the platform_data struct idiom. This gets rid of all the
flags fiddling.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Index: linux-2.6/drivers/scsi/aic7xxx/aic7xxx_osm.c
===================================================================
--- linux-2.6.orig/drivers/scsi/aic7xxx/aic7xxx_osm.c 2006-01-15 21:45:28.000000000 +0100
+++ linux-2.6/drivers/scsi/aic7xxx/aic7xxx_osm.c 2006-02-06 15:27:21.000000000 +0100
@@ -373,7 +373,6 @@
struct scb *);
static void ahc_linux_queue_cmd_complete(struct ahc_softc *ahc,
struct scsi_cmnd *cmd);
-static void ahc_linux_sem_timeout(u_long arg);
static void ahc_linux_freeze_simq(struct ahc_softc *ahc);
static void ahc_linux_release_simq(struct ahc_softc *ahc);
static int ahc_linux_queue_recovery_cmd(struct scsi_cmnd *cmd, scb_flag flag);
@@ -1193,7 +1192,6 @@
memset(ahc->platform_data, 0, sizeof(struct ahc_platform_data));
ahc->platform_data->irq = AHC_LINUX_NOIRQ;
ahc_lockinit(ahc);
- init_MUTEX_LOCKED(&ahc->platform_data->eh_sem);
ahc->seltime = (aic7xxx_seltime & 0x3) << 4;
ahc->seltime_b = (aic7xxx_seltime & 0x3) << 4;
if (aic7xxx_pci_parity == 0)
@@ -1830,10 +1828,9 @@
if (ahc_get_transaction_status(scb) == CAM_BDR_SENT
|| ahc_get_transaction_status(scb) == CAM_REQ_ABORTED)
ahc_set_transaction_status(scb, CAM_CMD_TIMEOUT);
- if ((ahc->platform_data->flags & AHC_UP_EH_SEMAPHORE) != 0) {
- ahc->platform_data->flags &= ~AHC_UP_EH_SEMAPHORE;
- up(&ahc->platform_data->eh_sem);
- }
+
+ if (ahc->platform_data->eh_done)
+ complete(ahc->platform_data->eh_done);
}
ahc_free_scb(ahc, scb);
@@ -2040,22 +2037,6 @@
}
static void
-ahc_linux_sem_timeout(u_long arg)
-{
- struct ahc_softc *ahc;
- u_long s;
-
- ahc = (struct ahc_softc *)arg;
-
- ahc_lock(ahc, &s);
- if ((ahc->platform_data->flags & AHC_UP_EH_SEMAPHORE) != 0) {
- ahc->platform_data->flags &= ~AHC_UP_EH_SEMAPHORE;
- up(&ahc->platform_data->eh_sem);
- }
- ahc_unlock(ahc, &s);
-}
-
-static void
ahc_linux_freeze_simq(struct ahc_softc *ahc)
{
unsigned long s;
@@ -2355,25 +2336,21 @@
if (paused)
ahc_unpause(ahc);
if (wait) {
- struct timer_list timer;
- int ret;
+ DECLARE_COMPLETION(done);
- ahc->platform_data->flags |= AHC_UP_EH_SEMAPHORE;
+ ahc->platform_data->eh_done = &done;
ahc_unlock(ahc, &flags);
- init_timer(&timer);
- timer.data = (u_long)ahc;
- timer.expires = jiffies + (5 * HZ);
- timer.function = ahc_linux_sem_timeout;
- add_timer(&timer);
printf("Recovery code sleeping\n");
- down(&ahc->platform_data->eh_sem);
- printf("Recovery code awake\n");
- ret = del_timer_sync(&timer);
- if (ret == 0) {
+ if (!wait_for_completion_timeout(&done, 5 * HZ)) {
+ ahc_lock(ahc, &flags);
+ ahc->platform_data->eh_done = NULL;
+ ahc_unlock(ahc, &flags);
+
printf("Timer Expired\n");
retval = FAILED;
}
+ printf("Recovery code awake\n");
} else
ahc_unlock(ahc, &flags);
return (retval);
Index: linux-2.6/drivers/scsi/aic7xxx/aic7xxx_osm.h
===================================================================
--- linux-2.6.orig/drivers/scsi/aic7xxx/aic7xxx_osm.h 2006-01-15 21:45:28.000000000 +0100
+++ linux-2.6/drivers/scsi/aic7xxx/aic7xxx_osm.h 2006-02-06 15:32:22.000000000 +0100
@@ -369,15 +369,12 @@
spinlock_t spin_lock;
u_int qfrozen;
- struct semaphore eh_sem;
+ struct completion *eh_done;
struct Scsi_Host *host; /* pointer to scsi host */
#define AHC_LINUX_NOIRQ ((uint32_t)~0)
uint32_t irq; /* IRQ for this adapter */
uint32_t bios_address;
uint32_t mem_busaddr; /* Mem Base Addr */
-
-#define AHC_UP_EH_SEMAPHORE 0x1
- uint32_t flags;
};
/************************** OS Utility Wrappers *******************************/
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH, RFC] aic7xxx: semaphore to completion conversion
@ 2006-02-06 18:54 Emmanuel Fusté
0 siblings, 0 replies; 3+ messages in thread
From: Emmanuel Fusté @ 2006-02-06 18:54 UTC (permalink / raw)
To: hch; +Cc: jejb, hare, linux-scsi, emmanuel.fuste
> On Tue, Jan 31, 2006 at 06:20:18PM +0100, Christoph Hellwig
wrote:
> > switch eh_sem to a completion. due to
wait_for_completion_timeout this
> > also nicely simplifies the code. Unfortunately it's
untested, so if
> > someone with the hardware could give it a try that would
be nice. Once
> > it works the same thing can be applied to aic79xx.
>
> New version that switches to the common onstack completion
and just a
> pointer in the platform_data struct idiom. This gets rid of
all the
> flags fiddling.
>
>
Tested and work.
But now (with this patch or the previous one) if I apply the
'Overlapped command error handling' fix, the kernel never
recover and a few second later completely freeze (keyboard
led, network activity ...).
It's like that with a queue depth of 2, one slot is stuck and
the driver continue to work with the second one (with big
delays like timeouts).
Will retry in console mode to try to catch somme messages.
Emmanuel.
Accédez au courrier électronique de La Poste : www.laposte.net ;
3615 LAPOSTENET (0,34 €/mn) ; tél : 08 92 68 13 50 (0,34€/mn)
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2006-02-06 18:55 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2006-01-31 17:20 [PATCH, RFC] aic7xxx: semaphore to completion conversion Christoph Hellwig
2006-02-06 14:40 ` Christoph Hellwig
-- strict thread matches above, loose matches on Subject: below --
2006-02-06 18:54 Emmanuel Fusté
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).